Daily Runoff Forecasting Using Ensemble Empirical Mode Decomposition and Long Short-Term Memory

Hydrological series data are non-stationary and nonlinear. However, certain data-driven forecasting methods assume that streamflow stable, which contradicts reality causes the simulated value to deviate from observed one. Ensemble empirical mode decomposition (EEMD) was employed in this study decompose runoff into several stationary components a trend. The long short-term memory (LSTM) model used build prediction for each sub-series. input set contained historical flow of simulation station, its upstream hydrological meteorological element series. final LSTM selected by MI method. To verify effect EEMD, Radial Basis Function (RBF) predict sub-series, decomposed EEMD. In addition, characteristics EEMD-LSTM different months runoff, GM(group month)-EEMD-LSTM up comparison. key difference between GM-EEMD-LSTM is GM must divide sequence on monthly basis, followed with EEMD model. results sub-series obtained RBF exhibited better statistical performance than those original series, especially EEMD-LSTM. overall low-water superior model, but flood season slightly lower both models significantly improved compared